blob: 3ae3f3bf2c1606b94d298a58489b16c078dcb4bd [file] [log] [blame]
# Copyright 2016 The Chromium Authors. All rights reserved.
# Use of this source code is governed by a BSD-style
# license that can be found in the LICENSE file or at
# https://developers.google.com/open-source/licenses/bsd
"""Implemention of the hotlist issues list output as a CSV file."""
from __future__ import print_function
from __future__ import division
from __future__ import absolute_import
from features import hotlistissues
from framework import framework_views
from framework import csv_helpers
from framework import permissions
from framework import xsrf
# TODO(jojwang): can be refactored even more, see similarities with
# IssueListCsv
class HotlistIssuesCsv(hotlistissues.HotlistIssues):
"""HotlistIssuesCsv provides to the user a list of issues as a CSV document.
Overrides the standard HotlistIssues servlet but uses a different EZT template
to provide the same content as the HotlistIssues only as CSV. Adds the HTTP
header to offer the result as a download.
"""
_PAGE_TEMPLATE = 'tracker/issue-list-csv.ezt'
def GatherPageData(self, mr):
if not mr.auth.user_id:
raise permissions.PermissionException(
'Anonymous users are not allowed to download hotlist CSV')
owner_id = mr.hotlist.owner_ids[0] # only one owner allowed
users_by_id = framework_views.MakeAllUserViews(
mr.cnxn, self.services.user,
[owner_id])
owner = users_by_id[owner_id]
# Try to validate XSRF by either user email or user ID.
try:
xsrf.ValidateToken(
mr.token, mr.auth.user_id,
'/u/%s/hotlists/%s.do' % (owner.email, mr.hotlist.name))
except xsrf.TokenIncorrect:
xsrf.ValidateToken(
mr.token, mr.auth.user_id,
'/u/%s/hotlists/%s.do' % (owner.user_id, mr.hotlist.name))
# Sets headers to allow the response to be downloaded.
self.content_type = 'text/csv; charset=UTF-8'
download_filename = 'hotlist_%d-issues.csv' % mr.hotlist_id
self.response.headers.add(
'Content-Disposition', 'attachment; filename=%s' % download_filename)
self.response.headers.add('X-Content-Type-Options', 'nosniff')
mr.ComputeColSpec(mr.hotlist)
mr.col_spec = csv_helpers.RewriteColspec(mr.col_spec)
page_data = hotlistissues.HotlistIssues.GatherPageData(self, mr)
return csv_helpers.ReformatRowsForCSV(
mr, page_data, '%d/csv' % mr.hotlist_id)